What Are Claude Connectors? Data Access, Auth, Custom Setup, and Troubleshooting (May 2026)

TL;DR: Claude Connectors are MCP servers that give Claude access to external data and services. The Connectors Directory has 375+ integrations for Google Drive, Slack, Notion, Adobe Creative Cloud, Spotify, Stripe, and more. You can add custom connectors via Settings. Auth uses OAuth when needed. Standard connectors return data; interactive connectors also render UI in the chat.

What Are Claude Connectors?

Claude Connectors are MCP servers that extend what Claude can do. Without connectors, Claude only has access to the conversation and any files you upload. With connectors, Claude can reach into your tools and services: search your Google Drive, read your Slack messages, query your Stripe dashboard, pull Jira tickets, generate designs in Figma, or access any API the connector exposes.

Each connector registers tools that Claude can call during a conversation. A Google Drive connector might expose a “search files” tool and a “read document” tool. A CRM connector might expose “find contact” and “update deal stage.” Claude decides when to call these tools based on your conversation, and the connector returns the data. Claude also suggests relevant connectors in context when it detects your conversation could benefit from one you haven’t enabled yet.

There are two types:

• Standard connectors return data that Claude formats into text. Claude decides how to present the information in its response.
• Interactive connectors also render UI components (cards, dashboards, forms) directly in the conversation using MCP App resources. First-party interactive connectors include Figma, Canva, Asana, Slack, Box, and Adobe Creative Cloud.

Both types are MCP servers under the hood. An interactive connector is a standard connector with a UI layer on top.

What Data Can Claude Access Through Connectors?

Claude accesses whatever data the connector’s tools expose. The connector author decides what’s available by defining tool schemas and handlers. The Connectors Directory has grown to 375+ integrations. Here’s what’s available across categories:

Communication and collaboration: Slack (messages, channels, users), Gmail (emails, drafts), Notion (pages, databases), Asana (tasks, projects), monday.com (boards, items)

Files and documents: Google Drive (documents, spreadsheets, presentations), Box (files, folders), DocuSign (documents, envelopes)

Development: GitHub (repos, issues, PRs, code search), Jira (issues, sprints, boards), Linear (issues, projects)

Sales and CRM: Salesforce (contacts, deals, accounts), Apollo (prospecting, contacts), Clay (enrichment), Outreach (sequences)

Analytics and data: Stripe (payments, customers, subscriptions), SimilarWeb (traffic, analytics), FactSet (financial data), Hex (data notebooks)

Design and creative tools: Figma (files, components, designs), Canva (designs, templates), Adobe Creative Cloud (50+ tools including Photoshop, Premiere, Express), Blender (3D modeling), Autodesk (CAD/engineering)

Music and audio: Ableton (music production), Splice (samples, sounds)

Consumer and lifestyle: Spotify, Uber, Instacart, Booking.com, Tripadvisor, AllTrails, Resy, StubHub, Taskrabbit, and more

Custom connectors can access anything: internal databases, REST APIs, GraphQL endpoints, proprietary services, file systems. If your tool handler can reach it, Claude can use it.

All data transfers are encrypted. Anthropic stores encrypted OAuth tokens but does not store the actual data your connectors return.

How Authentication Works

Whether a connector needs auth depends on what it does and how it’s deployed.

Connectors Directory connectors that access private data use OAuth. When you enable one for the first time, you’ll be redirected to the service (Google, Slack, Stripe, etc.) to authorize access. After that, Claude can call the connector’s tools using your authorized credentials.

Custom connectors you add via Settings don’t require auth during development. If your MCP server is running on localhost (exposed via ngrok), Claude connects directly. If your custom connector needs to protect its endpoints, you can configure OAuth Client ID and Secret in the advanced settings when adding it.

For Connectors Directory submission, if your connector requires authentication, you must use OAuth with user consent flow. Pure client credentials flow (machine-to-machine without user interaction) is not supported. You also need to provide a test account for Anthropic’s reviewers. Both https://claude.ai/api/mcp/auth_callback and https://claude.com/api/mcp/auth_callback must be allowlisted as redirect URIs.

Without OAuth, Claude does not pass any user identity information to your server. No user IDs, session tokens, or IP addresses. If you need per-user data segmentation, OAuth is the only mechanism Claude supports.

For a full walkthrough, see the Claude Connector OAuth guide.

How to Add a Custom Connector to Claude

1. Run your MCP server. If it’s local, expose it with a tunnel like ngrok: ngrok http 3000
2. Go to Claude Settings > Connectors
3. Click the plus icon > “Add custom connector”
4. Enter your MCP server URL with the /mcp path (e.g., https://abc123.ngrok.app/mcp)
5. Save

Your connector is now available in Claude conversations. Enable it per conversation via the + button in the message composer, or it may activate automatically when Claude determines it’s relevant.

Plan limits: Free Claude users can add one custom connector. Pro, Max, Team, and Enterprise plans support more. On Team and Enterprise plans, org admins must approve custom connectors first.

Claude Code and Connectors

Claude Code uses MCP servers directly rather than the Connectors UI. The same MCP server that powers a Claude Connector works in Claude Code, but the connection method is different.

In Claude Code, you configure MCP servers in your project settings or via the CLI. If you’re building a connector, the same server URL works in both Claude (via the Connectors UI) and Claude Code (via MCP settings).

Claude Managed Agents and Connectors

Claude Managed Agents, launched in April 2026, are a hosted runtime for long-running agent workflows. Managed Agents pair Claude’s agentic API with MCP servers, which means your Claude Connector’s tools are available to agents that run in the background, not just in conversations.

Managed Agents include OAuth flows, credential vaults, and managed compute. Since May 2026 they also support multiagent orchestration, where a lead agent delegates subtasks to specialist agents with their own models and tools. If you’ve already built an MCP server for the Connectors Directory, that same server can power a Managed Agent without changes.

Troubleshooting: “Could Not Load Connectors Directory”

This error shows up when Claude can’t reach the Connectors Directory. Common causes and fixes:

1. Browser extensions are the most common cause. Ad blockers, privacy extensions, and script blockers can interfere with the requests Claude makes to load the directory. Disable extensions or try incognito mode.
2. Network issues. Corporate firewalls, VPNs, or DNS problems can block the connection. Try a different network or disable your VPN temporarily.
3. Service disruption. Check status.claude.com for ongoing incidents.
4. Browser cache. Clear your cache and reload. A stale cached response can cause this.
5. Try a different browser. If it works in Chrome but not Firefox (or vice versa), the issue is browser-specific, likely an extension or setting.

If none of these work, contact Anthropic support through the help menu in Claude.

Building Your Own Claude Connector

A Claude Connector is an MCP server, so building one means building an MCP server with tools that connect to your data source. The server needs to use Streamable HTTP transport so Claude can reach it over the internet.

For a standard connector, you define tools with schemas and handlers that return structured data. Claude formats the data into text responses.

For an interactive connector, you also register MCP App resources: React components that Claude renders directly in the conversation. This is how connectors like Figma, Canva, and Adobe Creative Cloud display visual UI inside the chat.

Testing Your Connector Before Shipping

You can test a Claude Connector locally without a Claude account or API credits. sunpeak includes an Inspector that replicates the Claude runtime on localhost, so you can verify your tools, test authentication flows, and preview interactive UI without redeploying:

npx sunpeak inspect --server http://localhost:3000/mcp

The Inspector lets you toggle between Claude and ChatGPT runtimes, switch light/dark themes, and test different display modes. For automated testing, sunpeak also supports E2E tests, visual regression tests, and CI/CD pipelines via Playwright.

If you want to scaffold a new connector project from scratch:

npx sunpeak new

The Claude Connectors tutorial walks through building a connector end-to-end with full code examples.